GPU-Support for OSCAR

[nilsmelchert]

Hello,
I just started parallelizing OSCAR on the GPU. First tests with the simple cavity-scan, already showed pretty impressive results. The implementation can be seen at 6c94c74. It simply consists of a big multidimensional matrix multiplication, that takes place on the GPU. The current progress will take place on the gpu_support branch. Contributions are welcome and can be communicated here with me (@nilsmelchert) and @Jerome-LMA.

For the tests shown below, I ran the example script Example_cavity_FSR_scan.m with different resolutions:

• 64x64

• 128x128

• 256x256

• 512x512

The following hardware was used:

• GPU: NVIDIA GTX 1080

• CPU: i5-8600K (6-Cores)

[nilsmelchert]

One thing I should mention in this Issue, is that currently OSCAR works with 8 bit doubles. On graphics cards the floating precision is 4 bit. This means that we experience small losses of precision through discretization.
But in my eyes this is not critical. But I just wanted to mention it.